Aggressive Chess Openings, Measured: Which Ones Actually Make Club Opponents Go Wrong

Lists of aggressive chess openings are usually opinion. This one is measured. For 70 openings we counted how often Lichess opponents at each rating band go wrong in the first 15 moves, and how much each error is worth. Gambits top the table at every level, one opening leads all seven bands, and the result is not the usual list.

How do you measure whether an opening is aggressive?

We do not measure style. We measure what the opening does to the other side.

For each of 70 openings, the taught lines are walked as a tree of positions where the opponent is to move, cut at move 15. At each position, the Lichess games explorer supplies how often opponents in a rating band play each reply, aggregated across blitz, rapid and classical games. Each reply is graded by our own offline Stockfish 18 evaluation, the same lite build the app ships, at depth 20 with an eight-million-node ceiling. In this index a reply goes wrong when it concedes at least 40 centipawns against the book move, and it is a mistake at 100 centipawns or more. Those are grading floors for replies in the index, not the per-move labels the game analysis uses.

Two figures come out per opening and band:

• Opponents who go wrong: the probability that an opponent who reaches the opening concedes at least 40 centipawns somewhere on its taught lines by move 15.
• Win points per game: the winning chances, in percentage points, that those mistakes hand you per game reaching the opening, each weighted by how often it is played and how much it costs on the Lichess winning-chances curve; a forced mate scores the maximum of 50.

Seven rating bands are measured: Lichess under 1200, then each 200-point band up to 2200 and above; the product shows them on your own platform's scale as approximate conversions. This post quotes the Lichess 1600 to 1799 band unless it says otherwise, where coverage is strongest. The index was last rebuilt on 2026-09-08.

Which openings make club opponents go wrong most often?

The top ten at Lichess 1600 to 1799, ranked by win points per game. "Side" is the side you play; the opponent is the one making the errors.

The Budapest Gambit is first by a wide margin, and the reason is concentrated on one move. After 1.d4 Nf6 2.c4 e5, 23% of White players in this band push 3.d5 instead of taking the pawn, which the engine grades at 103 centipawns worse than the book move. Counting every wrong reply, 44% of opponents go wrong on move three alone, and half the opening's error mass has landed by then.

The King's Gambit is the opposite shape. Its opponents go wrong more often, six in ten by move 15, but the errors arrive later and are spread out. The single biggest is 4...Bg4 after 4.d4 in the Fischer Defense, played by 30% of Black players who reach that position, at 147 centipawns against the book move. Of the opponents still on the line at move five, 69% go wrong there.

Why do "how often" and "how much" disagree?

Because an opening can provoke many small errors or a few large ones, and the two figures measure different things.

The Scandinavian Defense has the highest error rate in the catalog, 69% of White players in this band going wrong by move 15, yet it is fifth by win points. Its errors are frequent and cheap; the largest single contributor, 2.Nf3, is played by only 5.6% of opponents. It is a source of small edges, not collapses.

The Traxler Counterattack is the mirror image. Only 3.6% of opponents go wrong at all, which puts it in the bottom third by error rate, yet it sits in the top fifteen by win points. The error that carries it comes after 5.Bxf7+ Ke7, where 12% of White players castle instead of retreating the bishop to d5, a mistake worth 347 centipawns. Rare, and close to decisive when it happens.

Does the punishment fade as ratings rise?

Yes, in total, and cleanly from the 1400 band upward. Summing win points per game across all 70 openings:

From the lowest band to the highest, the total falls by 71%. One caution on the low end: below 1400 the catalog is thinner. Only 41 of 70 openings register any win points under 1200, against 48 in the 1600s, and the share of replies carrying an engine verdict is lower there (66% against 74%), so part of what looks like less punishment down there is less measurement. The median error rate is flat through the 1600s and only falls from the 1800s.

What does not fade is the head of the table. The Budapest Gambit is first in all seven bands and at 2200 and above still hands Black 1.11 win points per game, with 21% of opponents going wrong; the King's Gambit still sees 40% of Black players go wrong. The decline is in the long tail of quiet openings, whose small edges disappear as opponents get stronger.

Which gambits actually work at club level?

Thirteen openings in the catalog are gambits by our own classification. Here they are at Lichess 1600 to 1799, with the two figures side by side:

Three things stand out. The Scotch Gambit provokes the highest error rate of any gambit, 63%, but most of its errors fall under the 100-centipawn line, so it converts to half the King's Gambit's win points. The Benko Gambit is the extreme case: six in ten opponents go wrong and none of it counts as win points, because every measured error is under the 100-centipawn line. The Evans and Smith-Morra, two well-known attacking gambits, convert to under a tenth of the King's Gambit's win points at this band. Opponents have seen them.

Why is every number a lower bound?

Because rare replies are not counted. Replies played in fewer than ten games or under 2% of games at a position are dropped, and only the six to ten most common replies per band are kept; a pruned reply can only add errors, never remove one. The walk also covers only the lines we teach, to 22 plies, so an opponent who leaves theory into a line we do not teach is not scored.

That is why a zero in these tables means "unmeasured", not "safe". The Queen's Gambit Accepted, the Vienna Game and the Torre Attack read 0.00 win points at every band. Each row also carries a coverage figure, the share of non-book replies with an engine verdict; a low one is a reason to distrust the row, not to call the opening harmless.

How should you use this to pick an opening?

Start from reach: win points per game assumes the opening appears, so check how often your opponents let it, which your own game history tells you. Then check the median move, the move by which half the errors have arrived: three for the Budapest, eight for the Two Knights. Punishing on move three is easy to prepare; move eight asks more of your memory.

None of this replaces knowing the punish. An opponent who plays 3.d5 against the Budapest has handed you a pawn's worth of evaluation only if you know the continuation, which is what learning an opening properly is for. In the product, the Explore openings board under Train sorts by this index and each row opens that opening's study.

Every number here can be regenerated from a clean checkout of the repository. The model is in src/lib/train/bounty-score.ts, the generator is scripts/gen-bounty-index.ts, and npm run build:bounty rebuilds the index in seconds from the committed reply data.

FAQ

Does an aggressive opening still work against stronger players?

The sharp ones do. Total punishment across 70 openings falls 71% from the lowest Lichess band to 2200 and above, but the Budapest Gambit is first in every band and still worth 1.11 win points per game at the top. The quiet openings are the ones whose small edges vanish.

Are these numbers from Chess.com games?

No. They are aggregated from the Lichess games explorer, blitz, rapid and classical, by Lichess rating bucket, with no individual games or players involved. The product converts the bands to your own platform's scale; how the two rating scales relate is covered separately.

Why does the Scandinavian rank so high if it is not aggressive?

Because the table measures opponent errors, not attacking style. Sixty-nine percent of White players at Lichess 1600 to 1799 go wrong against it by move 15, the highest rate in the catalog. The errors are small, which is why it is fifth by win points rather than first.

Does going wrong mean the opponent blundered?

Not necessarily. In this index, "go wrong" means a reply that concedes at least 40 centipawns against the book move at depth 20. Win points count only replies at 100 centipawns or more, weighted by the winning chances they cost; a high error rate with few win points means small errors.

Can I see these figures for my own rating?

Yes. The Explore openings board under Train in AxiomChess has a chip for each rating band on your own platform's scale and sorts the catalog by this index for the band you choose, with the Lichess game count in the caption.

Pick the opening your opponents get wrong, then learn the punish. Analyze your games free to see which openings actually appear in your games, and which of them your opponents are already misplaying.